📌 Quick Summary: Stop struggling with slow local models! Use Claude Code completely free via NVIDIA NIM API - no GPU required. Get 40 requests per minute (no daily limit). Works on any laptop. This step-by-step guide covers Mac, Linux, and Windows (WSL). Build real projects, not just demos. 100% free, no hidden costs.
🤖 Claude Code for Free - Using NVIDIA NIM API (No GPU, No Subscription)
Hey everyone! Welcome back to PakAlerts. Today I'm going to show you something that completely changed how I code. You've probably seen videos where people tell you to install Ollama and run local models. And maybe you tried it. But let's be honest - unless you own a $3,000+ workstation with a high‑end GPU, those local models are painfully slow. Even on a powerful MacBook Pro, running an intelligent model locally makes coding impossible.
So what's the solution? Cloud APIs. And recently, NVIDIA did something amazing - they launched their NIM API and made several powerful models completely free. We're talking about models like DeepSeek V4 Pro, GLM 5.1, Qwen, MiniMax - all comparable to Claude Sonnet and Opus. And the best part? No daily token limit. The only limit is 40 requests per minute, which is more than enough for serious development work.
In this guide, I'll walk you through the entire setup - from getting your free NVIDIA NIM API key to running Claude Code on Mac, Linux, or Windows (using WSL). I'll also show you a live demo of building a functional web app. No fluff, no copy‑paste from other tutorials - just what actually works.
📢 Before we start: All the commands and links mentioned in this guide are also available on our website - you can copy them directly. No need to type everything manually. I'll keep this article open as we go through each step.
🧰 What You'll Need
- A computer (Mac, Windows, or Linux - old laptops work fine).
- A free NVIDIA account (to get API keys).
- A terminal (iTerm, Terminal, PowerShell, or Ubuntu on WSL).
- About 20 minutes of focused time.
🎯 What We're Building
By the end of this guide, you'll have Claude Code running with NVIDIA's free models. You'll be able to:
- Ask Claude Code to build full‑stack applications.
- Create websites, APIs, and automation scripts - all for free.
- Work without worrying about token limits or expensive hardware.
📥 Step 1: Install uv and Python (One‑time setup)
uv is a fast Python package manager. We need it to run the NVIDIA NIM proxy server.
On Mac or Linux:
Open your terminal. Copy and paste this command:
curl -LsSf https://astral.sh/uv/install.sh | sh
Then install Python 3.14:
uv python install 3.14
On Windows (via WSL - the proper way):
First, open PowerShell as Administrator and run:
wsl --install
Restart your PC when asked. After reboot, search for "Ubuntu" in the Microsoft Store, install it, and launch it. Set a username and password - this is your Linux environment inside Windows.
Inside Ubuntu, copy and paste the same Mac/Linux commands above - they work perfectly.
💡 Why WSL? Most AI tools are built for Linux first. WSL gives you a real Linux kernel on Windows, making everything much smoother. Don't use PowerShell for this - you'll run into issues.
📥 Step 2: Download the NVIDIA NIM Proxy Server
Create a folder where you'll keep the proxy server. I call mine nvidia-nim. Run these commands:
mkdir nvidia-nim cd nvidia-nim git clone https://github.com/Alishahryar1/free-claude-code.git . cp .env.example .env
The last command copies the example environment file. Now open this folder in VS Code (code .) or any text editor. You'll see a .env file - we'll edit it in a moment.
🔑 Step 3: Get Your Free NVIDIA NIM API Key
- Go to NVIDIA NIM API Keys page. Sign up or log in with Google/GitHub.
- Verify your account using your phone number - this is mandatory to generate keys.
- Click Generate API Key. Give it a name like "claude-code" and set expiration to "Never Expire".
- Copy the key - you'll see it only once.
- Paste it into the
.envfile underNVIDIA_API_KEY=.
Save the file. That's it - your key is ready.
🧠 Step 4: Choose Your Model (DeepSeek, GLM, Qwen, etc.)
Go to NVIDIA NIM Models page. Click the "Free Endpoint" filter to see all free models. I recommend:
- DeepSeek V4 Pro - excellent for coding, very smart.
- GLM 5.1 - great all‑rounder.
- Qwen 2.5 - fast and reliable.
- MiniMax - comparable to Claude Sonnet.
Click on your chosen model, then click View Code. Look for the string like nvidia/deepseek-v4-pro - copy that.
Open the .env file again. Find NVIDIA_MODEL= and paste it there. Keep the nvidia/ prefix - don't remove it.
🚀 Step 5: Run the NVIDIA NIM Server
Go back to your terminal (inside the nvidia-nim folder). Run:
uv run uvicorn server:app --host 0.0.0.0 --port 8082
This will download a few dependencies and start a local API server. Keep this terminal window running - Claude Code will send requests here. You'll see output like "Uvicorn running on http://localhost:8082". That's good.
📥 Step 6: Install Claude Code
Open a new terminal window (don't close the previous one). Run:
curl -fsSL https://claude.ai/install.sh | bash
After installation, restart your terminal or open a new tab. Then navigate to the folder where you want to create your project. For example:
mkdir my-awesome-project cd my-awesome-project
⚡ Step 7: Launch Claude Code with NVIDIA NIM
Now the magic command - this tells Claude Code to use your local NVIDIA proxy server instead of Anthropic's paid API.
On Mac / Linux / WSL (Bash):
ANTHROPIC_AUTH_TOKEN="freecc" ANTHROPIC_BASE_URL="http://localhost:8082" claude
That's a single line. Press Enter. Claude Code will start. Choose your theme (Dark / Light) and press Enter again.
If everything is correct, you'll see a status line. Type /status - you should see the localhost URL, confirming it's connected to your free NVIDIA model.
📌 Pro tip: To avoid typing that long command every time, create an alias. Add this to your ~/.zshrc or ~/.bashrc:
alias claude-free='ANTHROPIC_AUTH_TOKEN="freecc" ANTHROPIC_BASE_URL="http://localhost:8082" claude'
Then just type claude-free to launch.
🧪 Step 8: Test It - Quick Demo
Inside Claude Code, type:
Explain HTML in one line
You'll see the API call happening in your other terminal window (the NVIDIA server). Within seconds, Claude Code will answer. That means your setup is 100% working!
🛠️ Live Project: Build a Task Manager with Dark Mode
Now let's do something real. I'll create a new folder and ask Claude Code to build a full task manager app.
I gave Claude Code this prompt:
"Create a task manager web app with dark mode. Users can add, complete, and delete tasks. Store tasks in local storage. Use modern HTML, CSS, and JavaScript."
Claude Code analyzed the request, created multiple files (HTML, CSS, JS), and started a local server. Within about two minutes, I had a fully functional task manager app - with a dark mode toggle, "Active" and "Done" tabs, and a "Clear all" button. No bugs. It worked flawlessly.
You can see - these free NVIDIA models are incredibly powerful. And I didn't spend a single rupee.
⚠️ Important Notes & Tips
- Keep the NVIDIA server running - whenever you want to use Claude Code, first run the
uv run uvicorn...command in a terminal, then launch Claude Code with the environment variables. - Rate limit - 40 requests per minute is generous, but if you hit it, just wait a few seconds. For normal coding, you'll rarely exceed it.
- If NVIDIA NIM is slow during peak hours - try switching to a different model (e.g., from DeepSeek to GLM). Or use the OpenRouter alternative (I have a separate guide on this).
- Windows users - always work inside Ubuntu on WSL. Don't use Command Prompt or PowerShell for this - the proxy server and Claude Code are designed for Linux. WSL gives you exactly that.
- Use Git - commit your code before asking Claude Code to make big changes. That way you can revert if something goes wrong.
✅ Sandbox mode for safety: Inside Claude Code, type /sandbox and choose "Sandbox with auto‑allow". This restricts Claude Code to only your project folder - it can't mess up system files. It will still auto‑accept edits, but you stay safe.
❓ Frequently Asked Questions
Q: Do I need a powerful graphics card?
A: No. Everything runs in the cloud. Your laptop just sends text - even a 10‑year‑old machine works fine.
Q: Is NVIDIA NIM really free? Any hidden costs?
A: Yes, it's genuinely free. No credit card required. The 40 requests per minute limit is soft - they don't cut you off. It's a promotional free tier from NVIDIA.
Q: Can I use this for commercial projects?
A: Absolutely. You own the code you generate. NVIDIA's terms allow commercial use.
Q: What if I need more than 40 requests per minute?
A: That's 2,400 requests per hour. Unless you're running massive automation, you won't hit it. If you do, you can create multiple NVIDIA accounts or use OpenRouter as a backup.
Q: I'm stuck. Where can I get help?
A: Leave a comment below this article - I personally reply to every query. You can also join our WhatsApp channel (link below) for quick help.
🚀 Next Steps & What to Learn
Now that you have Claude Code running for free, here's what you can do next:
- Watch our complete Claude Code tutorial (link in related posts) to learn all the features - YOLO mode, plan mode, MCP, skills, and token saving tricks.
- Build your first full‑stack application - try a blog, a portfolio site, or a small e‑commerce page.
- Explore other free APIs - OpenRouter and DeepSeek also provide free models that you can integrate.
📢 Final Words
I've been using this setup for weeks, and it's changed how I code. No more worrying about API costs or slow local models. NVIDIA has given us a true gift - use it wisely.
If this guide helped you, share it with a friend who's struggling with expensive AI tools. And if you have any questions, don't hesitate to ask. I'm here to help.
🔗 Resources (all links are on this article for easy copy): NVIDIA NIM API Keys, NVIDIA Models page, uv installation docs, free-claude-code GitHub repo - use the commands above or visit the official sites.
⚠️ Disclaimer: This method uses third‑party APIs. Always respect the terms of service. NVIDIA may change their free tier in the future - but as of April 2026, this works perfectly.
